Guide · WhatsApp and AI assistants

WhatsApp in your AI assistant’s context.

“Who still needs a reply?”, “Who did I speak to often but hear from less now?”, “Draft a short response”. Gravify gives your chosen agent access to your WhatsApp conversations and their history.

Turn a list of chats into a useful question

Gravify gathers available history on your computer and connects conversations to people. Through MCP, a compatible assistant can read authorized context, search messages and prepare drafts. Start with one chat or ask for an overview.

  • Find replies you still owe. Ask which conversations have an incoming message after your latest response, and how long they have been waiting.
  • Prepare a draft. Have the agent read the conversation and specify tone, length and key points: “Look at how I write to this person. Two lines, without emoji”.
  • See how contact changes. Compare two periods to find people you interact with more or less.
  • Reconnect with someone. Start with the people you want to keep in touch with and ask who is missing from recent interactions.

Answers depend on imported history and the data available to the assistant. An unread chat and an unanswered conversation are different; urgency also requires reading the content. Frequency describes observed interactions, not how much a person matters to you.

Claude, ChatGPT or Codex: which connection?

On Windows, Gravify provides local setup for Codex and Claude Code in Settings → Assistant & agents. Other local MCP clients can use the Personal Hub URL and authorization header with Streamable HTTP.

If you use ChatGPT, the supported local route here is Codex, with its own sign-in and your account’s limits. Claude Code is the local route for Claude. ChatGPT and Claude web chats and cloud-hosted connectors cannot directly reach Gravify’s local address: this guide is for an agent running on your computer.

The Personal Hub keeps data on hardware you control. Context requested by the agent is shared with the model you choose. In Gravify, you can choose which data is accessible and revoke a connection.

Let your agent set up Gravify

An agent that can operate your computer can handle installation and configuration. Copy the prompt below into Codex, Claude Code or your local assistant. It starts by checking what is already installed and guides you through the personal steps.

  1. Install Gravify 0.2.0 or later for Windows from the official website and activate your licence. If this version is available only on the Tester channel, check its release notes before choosing it.
  2. Open Settings → Accounts & Data Sources → WhatsApp, read the notice for the optional unofficial connector, choose a retention mode and link your phone. Reading chat text requires full local history; metadata mode supports interaction analysis.
  3. In Settings → Assistant & agents, choose the data your assistant can access and configure your local agent.
  4. Restart the agent client and try a read. Initial history import may still be in progress.

To keep MCP available when the window is closed, enable Run Gravify in the background and choose Always Ready in the background behavior settings. The computer and Personal Hub must remain running.

Three requests to try after connecting

Look at the available WhatsApp chats: which ones are waiting for my reply? Show the messages you rely on before suggesting a priority order.
For this person, prepare a short draft in the style of my previous messages. If you need a decision from me, ask first.
Compare this month’s interactions with the previous month: who am I in touch with less? Distinguish a real decline from incomplete history.

You can review drafts in Gravify, edit them and, for direct chats, open the conversation in WhatsApp. To explore a trend, choose periods with enough available history.

Talk to your assistant through WhatsApp.

In the Tester channel, set up a dedicated assistant number and message it from your verified personal number. Local setup guides you through pairing and the first reply. This optional, unofficial integration requires risk acceptance and local history storage. Your Hub must stay running; the number and AI service have separate costs. It replies only to you, with no groups, other recipients or proactive messages.

This account is separate from the connector that imports your conversations. Drafts for other people remain subject to approval and manual sending.